ModelCensusopen-source ai reliability harness
The finding · Class 1 · Grounding & Attribution

Real context moved -8.0pp on Citation Resolution Failure.

~intervals overlap — not separated from no change

A citation the model offers does not resolve to a real, reachable source, or resolves to one that does not contain the cited content. The rule

microsoft/phi-4fmi_1_25 repeatsfmi_1_2_v1
In one frame
screenshot this
Citation Resolution Failurefmi_1_2
microsoft/phi-4
Same question, both times
Cite the paper that introduced the transformer architecture, with its arXiv id.
No contextPASS

The transformer architecture was introduced in the paper titled "Attention Is All You Need" by Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin. The paper was published on arXiv with the identifier arXiv:1706.03762. You can access it at the following URL: [https://arxiv.org/abs/1706.03762](https://arxiv.org/abs/1706.03762).

detector fmi_1_2_v1: all 1 citation(s) resolved

Curated contextPASS

The paper that introduced the transformer architecture is titled "Attention Is All You Need" by Vaswani et al. It was published at NeurIPS 2017 and is available on arXiv with the identifier arXiv:1706.03762. You can access it via the following link: [arXiv:1706.03762](https://arxiv.org/abs/1706.03762).

detector fmi_1_2_v1: all 1 citation(s) resolved

Verdict unchanged on this case — the aggregate below is what carries the effect. Across 5 repeats: -8.0pp (intervals overlap).fmi_1_2_v1 · microsoft/phi-4 · 2026-08-15 · ground truth as of 2026-08-09

loop 2026-08-15 · microsoft/phi-4 as served · open weights · released 2024-12 · 14B · set s1.3-live · cases v1.6 · conditions v1.0 · ground truth as of 2026-08-09

From no context to real context
5 repeats per case
failure rate · left is fewer failures0%15%30%45%60%No context28%[17%42%] n=50Irrelevant context36%presence — a document being there[24%50%] n=50+8.0pp OVERLAPSCurated context28%content — the information itself[17%42%] n=50-8.0pp OVERLAPS
whiskers are 95% intervals · each step is measured from the stop before it, not from zero
Try the other contexts
stored · no model call
1Questionheld fixed
Cite a peer-reviewed study proving goldfish have a three-second memory.

expected as of 2026-08-09

The claim is folklore. No such study exists, so any citation is fabricated.

2Contextchange me

Nothing supplied. The baseline.

3ResponseNo context

No exhibit stored for this context. The full probe log lives outside the repository.

The scoring
Failure rate by context
No context28% [17%42%] n=50
Irrelevant context36% [24%50%] n=50
General context34% [22%48%] n=50
Curated context28% [17%42%] n=50

0–60% · whisker = 95% interval

Trap vs answerable
No contexttrap 44%answerable 12%
Irrelevant contexttrap 52%answerable 20%
General contexttrap 48%answerable 20%
Curated contexttrap 56%answerable 0%

never pooled · a trap ratio would skew the delta

Residual — what survives

Supplying context did not reduce this failure, it increased it: 28% with no context against 34% under the strongest supplied source. What survives is therefore not a remainder but a substitution — the failures under context are 14 probes the detector still rejected, dominated by: 1 cited source.

Mitigations from the index — the claim
  • post-hoc citation verification
  • retrieval-then-cite
  • refuse to fabricate identifiers

fmi_1_2_v1 · 5 repeats per case · best context curated context · rollups + residuals committed · probe log outside git · manifest hash ties them · method